On Mon, Oct 26, 2020 at 08:37:00PM +0800, Hanjun Guo wrote:
[...]
in preparation for the discussion, may I ask you please to elaborate a bit more on the Uncore DVFS topic so that we have more background info ?
Sure, for the later generation of Kunpeng server chip, there is a Uncore block in the chip which contains L3 cache, ring buses and etc., the Uncore stays in a single power domain and can do DVFS at runtime. As the Uncore consumes considerable power, so it's better to control the frequency at runtime.
The Uncore can be abstracted as a power device, and the power can be controlled via a single interface such as MMIO, but the permission is only restricted in EL3 (avoid spectre via DVFS), so I think we need a spec update to control the power for Uncore devices, like PSCI or SCMI, or an common interface for ACPI spec.
Please let me know if I didn't explain it in detail, I can prepare a slide for the meeting if needed.
You did explain it properly and yes I would encourage all people who raised a topic to prepare a slide or two to start the technical session.
Please everyone note that these sessions are there to discuss ongoing work technical details and how to make progress on those, not patches upstreaming plans.
Thank you very much.
Lorenzo